Pitch Estimation via Precomputed Lookup Tables

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional pitch-estimation technologies are inefficient in estimating the weight of the probability density function of a fundamental frequency and relative amplitude of harmonic components in real-world sound mixtures, requiring extensive computations and being unsuitable for sound mixtures with overlapping frequency components and no fundamental frequency components.

Innovation Solution

A pitch-estimation method that represents observed frequency components as a probability density function, uses multiple tone models, tone model parameter estimation, and introduces prior distributions for model parameters, reducing computations by storing precomputed values and discretizing the frequency axis, thereby estimating the weight and relative amplitude of harmonic components more efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional pitch-estimation technologies are used to estimate the weight of the probability density function of a fundamental frequency and relative amplitude of harmonic components in real-world sound mixtures, then the estimation can be performed, but extensive computations are required and the technology is unsuitable for sound mixtures with overlapping frequency components and no fundamental frequency components

Engineering Contradiction:
Improveestimation accuracyVSAvoidcomputational time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-calculating and storing values of the function g(γ) = exp(-γ²/2σ²) in a lookup table before the main estimation process. This precomputation stores frequently needed exponential values that would otherwise require repeated calculation during the EM algorithm iterations, significantly reducing the computational burden while maintaining estimation accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent transforms the continuous frequency domain into a discrete domain by introducing a discretization parameter δf. The continuous probability density function is converted into a discrete representation where frequencies are sampled at intervals of δf. This parameter change from continuous to discrete domain enables efficient computation using digital signal processing techniques while preserving the essential characteristics of the pitch estimation.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If conventional pitch-estimation technologies assume a small number of sound sources and locally trace frequency components or depend on existence of fundamental frequency components, then the computation is simpler, but the technology cannot be applied to real-world sound mixtures where the number of sound sources is unknown and frequency components overlap

Engineering Contradiction:
Improveapplicability to real-world sound mixturesVSAvoidcomputational complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent creates a universal pitch estimation method that works for any sound mixture regardless of the number of sound sources or presence of fundamental frequency components. The method uses a mixture model that can represent any combination of harmonic and non-harmonic sounds through the probability density function p(f; θ), where θ includes fundamental frequencies, amplitudes, and spectral shapes. This universal approach eliminates the need for assumptions about sound source characteristics.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent introduces an intermediary statistical model (mixture distribution model) that bridges the gap between observed frequency components and underlying sound sources. Instead of directly tracing frequency components or assuming fundamental frequencies exist, the method uses probability density functions as intermediaries to represent and analyze the sound mixture, enabling handling of complex overlapping scenarios.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If multiple tone models and tone model parameter estimation are used to improve pitch estimation accuracy, then the estimation can handle complex sound mixtures, but the computational load increases significantly

Engineering Contradiction:
Improvepitch estimation accuracyVSAvoidcomputational efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent applies preliminary action by pre-computing and storing the function g(γ) values in a lookup table. During the EM algorithm iterations with multiple tone models, instead of calculating exp(-γ²/2σ²) repeatedly, the system retrieves precomputed values from the table, reducing computational overhead while maintaining the accuracy benefits of multiple tone models.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses approximation by copying precomputed values from the lookup table rather than performing exact calculations each time. The discrete sampling of the probability density function at frequency intervals δf creates a copied representation that approximates the continuous function, enabling efficient computation with multiple tone models while preserving essential accuracy.

Inventive Principle:
Principle #26Copying

Data Source

PatentUS7885808B2Pitch-estimation method and system, and pitch-estimation program
Publication Date: 2011.02.08 NATIONAL INSTITUTE OF ADVANCED INDUSTRIAL SCIENCE & TECHNOLOGY
  • US7885808B2 patent drawing
  • US7885808B2 patent drawing
  • US7885808B2 patent drawing

AI summary

A pitch-estimation method, a pitch-estimation system, and a pitch-estimation program are provided, which estimate a weight of a probability density function of a fundamental frequency and relative amplitude of a harmonic component through fewer computations than ever. In the improved pitch-estimation method, 1200 log2 h and exp[−(x−(F+1200 log2 h))2/2W2] in the following expression are computed in advance and then stored in a memory of a computer:c′⁡(t)⁡(h❘F,m)⁢12⁢π⁢⁢W2⁢exp(-(x-(F+1200⁢⁢log2⁢h))22⁢W2)(61)The above expression is computed only with respect to a fundamental frequency F wherein x−(F+1200 log2 h) is close to zero. With this arrangement, computations to be performed may considerably be reduced, and computing time may accordingly be shortened.